查看文章

View Video Presentation: https://doi.org/10.2514/6.2023-1435.vid

We assess the capability of recurrent neural networks in predicting unsteady forces on moving structures in fluids. We consider the case of a plunging oscillating flat plate and aim at predicting the unsteady lift force using relatively short-time of few cases having different excitation amplitudes. The approach is based on predicting temporal coefficients of modes generated by applying proper-orthogonal decomposition to the ensemble data. This information is then applied to train neural networks which is capable of predicting POD dynamics over a broad range of oscillating amplitudes. In addition to open-loop neural network, a closed-loop network, using long-short term network, is applied to develop a reduced order model of lift coefficient through integration of pressure modes. The results are validated with those obtained from UVLM simulations.